How to Edit an Image with a Text Prompt

Prompt-based image editing works best when you treat the source image as a list of decisions to preserve. A short request such as “make it better” leaves the model to decide what should change. A controlled prompt names the edit, the protected details, and the desired finish.

Start with one clear change

Upload the strongest version of your source image, then describe the highest-priority change first. Useful requests include replacing a background, changing a garment color, removing an object, correcting lighting, or adapting the composition to a new aspect ratio.

For example:

Replace the studio background with a bright modern kitchen. Keep the person, pose, clothing, facial features, and camera angle unchanged. Match the original direction of light and preserve realistic shadows.

This prompt separates the variable from the constants. It also gives the model visual constraints that can be checked in the result.

Say what must stay unchanged

Identity, product geometry, logos, text, and small surface details can drift during an edit. List the details that matter to the job. For a product photo, that may be the package proportions and label layout. For a portrait, it may be facial identity, hairstyle, and pose.

Avoid an exhaustive list of every pixel. Three to five protected details usually give the instruction enough focus.

Describe the finish in visual terms

Words such as “professional” are subjective. Add observable choices: soft window light, clean white backdrop, eye-level camera, natural skin texture, shallow depth of field, or restrained contrast. If the output has a commercial purpose, name the format too: marketplace thumbnail, vertical social post, editorial cover, or website banner.

Iterate from the best result

When the first edit is close, continue from that image and ask for one correction. Repeating the entire creative brief can introduce changes you did not request. A follow-up such as “keep everything else unchanged; reduce the shadow under the product and center the label” is easier to evaluate.

Each edit consumes generation resources, so focused revisions also control cost. Check the preview at full size before starting another run, then download the approved result or keep it in your library for later work.

A reusable prompt structure

Use this order when a task needs precision:

1. Change: the single visual edit you want.
2. Preserve: identity, objects, text, pose, composition, or branding that must remain.
3. Match: lighting, perspective, shadows, and texture that should stay coherent.
4. Deliver: aspect ratio and intended use.

Open the image workspace to try the structure, or review plans and credits before a larger batch.